Abnormal detection method of flotation dosing based on nsst morphological features and depth kelm

Abstract

The invention relates to a flotation dosing abnormality detection method based on NSST morphological features and depth KELM. Firstly, the bubble image on the surface of the flotation cell is collected in real time, and the NSST image is decomposed to obtain the low-frequency sub-band image and the multi-scale high-frequency sub-band; secondly, the low-frequency image is binarized to extract bubble bright spots, and the number and area of ​​each bright spot are calculated , standard deviation and ellipticity, and calculate the fractal dimension, mean and variance of the high-frequency subband coefficients of each scale to form the multi-scale morphological features of the bubble image; then, on the basis of the KELM algorithm, a deep KELM is constructed with deep learning , the quantum computing is introduced into the optimization of the genetic algorithm, and used to optimize the parameters of the depth KELM, and construct the adaptive depth KELM; finally, establish the flotation dosing anomaly detection model through the multi-scale morphological features and the adaptive depth KELM. The average recognition rate and operating efficiency of the present invention are significantly higher than the existing detection methods, more in line with the requirements of on-line detection in flotation production, and lay a foundation for the subsequent automatic control of dosing.

Description

technical field [0001] The invention relates to a flotation dosing abnormality detection method based on NSST morphological features and depth KELM. Background technique [0002] In the mineral flotation process, the flotation agent is one of the most critical control quantities. The quality of the dosage directly affects the mineral processing production indicators. relevant. When the dose is normal, the size of the bubbles is moderate, the size distribution is uniform, and the circularity of the bubbles is high; when the dose is over, the bubbles are severely hydrated and have strong fluidity, mainly small-sized bubbles; Higher, the circularity of the bubbles is low, and a large number of bubbles merge. At present, the concentrator mainly adopts artificial naked eyes to observe the changes of the characteristics of the air bubbles on the surface of the flotation tank to adjust the dosage, and the judgment and control lag, and the subjective randomness is large. [0003]...

AI Technical Summary

Problems solved by technology

However, the input weights and hidden layer bias of ELM are randomly selected, the number of hidden layer nodes is difficult to determine, and problems such as overfitting will directly affect its stability and generalization ability.

Therefore, Huang et al. introduced the kernel function into the ELM algorithm and proposed the kernel extreme learning machine (KELM), which enhanced the generalization performance of the algorithm, but at the same time its performance was easily affected by the penalty coefficient C and the kernel function σ

Images